Unexpected powerlevels interactions

We accept some PL event transitions that are maybe unexpected.

Suppose we have a room where:
* `events."m.room.power_levels"` has been lowered to 50 (ie, "moderators" have been allowed to change the power levels)
* `users."@mischievous_mod:example.com"` is 50 (ie, they are a moderator)
* `events_default` is 75, so `@mischievous_mod` (and other users) cannot send arbitrary events.

Now:

1. `@mischievous_mod` can add new entries to `events`, thus allowing themselves (or other users) to send events. This appears to be counter to the spirit of point 10.4.1 of the [auth rules](https://spec.matrix.org/v1.4/rooms/v1/#authorization-rules):

> For each entry being added ... in both the events and users keys: if the current value is higher than the `sender`’s current power level, reject.

... which seems to seek to prevent the sender of a PL event from controlling the transmission of events that they themselves do not have permission to send.

2. Conversely, suppose there is an existing entry in `events` setting `"m.room.message": 0`. The mischievous moderator can *remove* this entry, thus changing the required PL for sending such an event to the default (75). This appears to be counter to the spirit of point 10.4.2:

> If the new value is higher than the `sender`’s current power level, reject.

... which seems to seek to prevent the sender of a PL event from preventing other users with the same PL from sending that event.

Suppose now that `users_default` is 75.

3. `@mischievous_mod` can add an entry to the `users` key *demoting* another user with a higher power level than themselves.
4. `@mischievous_mod` can *remove* an entry from the `users` key, *promoting* another user (or themselves) to a higher power level than their own.

Start with the authorization rules section linked in the issue and compare its wording with the four power-level transitions described here. Determine whether these behaviors are intended or require a specification change; done requires a decided interpretation and an agreed update to the Matrix specification.

Written by the indexing model from the issue text.